This study aims to improve the reliability, accuracy and automation of inkjet-based three-dimensional (3D) printing by developing an intelligent and real-time printhead status detection method. The proposed approach is applicable to all types of inks, with a particular focus on the challenges of detecting nozzle failures when using colorless transparent inks – issues that are difficult to assess using conventional monitoring techniques.
A low-cost, highly adaptable machine vision system was developed to capture test patterns printed with various types of inks, including transparent ones. A pixel weighting and multifeature fusion (PMF) algorithm was introduced to evaluate nozzle health by integrating pixel intensity, centroid deviation and connected component analysis. In addition, a data set comprising 10,800 labeled images was constructed, and three lightweight convolutional neural networks (LeNet, VGG-Small and AlexNet-Small) were trained to classify nozzle states into three categories: normal, intermittent and clogged.
The PMF algorithm effectively provides a quantitative assessment of nozzle and overall printhead condition without requiring data-driven training. Meanwhile, the CNN models demonstrated strong generalization and robustness in identifying various types of nozzle defects. All models achieved classification accuracy exceeding 99.5% and enabled real-time evaluation at speeds over 4,100 nozzles/s.
This study proposes a novel hybrid detection framework that combines interpretable image features with deep learning-based classification. It significantly enhances the current capabilities of printhead condition monitoring. The proposed system improves automation and operational reliability in inkjet-based 3D printing and provides a practical, scalable solution for intelligent maintenance in industrial environments.
